If you're using Julia and want something like this, you can try `LoopVectorization.vfilter`. It lowers to LLVM intrinsics that should do the right thing given A

If you're using Julia and want something like this, you can try `LoopVectorization.vfilter`. It lowers to LLVM intrinsics that should do the right thing given AVX512, but performance is less than stellar without it IIRC. May be worth taking another lookat. I'm not licking this cookie; I welcome anyone else wanting to get their hands dirty.